Chinese smartphone and electric car manufacturer Xiaomi has unveiled the first in-house development of the MiMo artificial intelligence language model.
This model is open source and includes 7 billion different parameters. According to the results of performance tests, MiMo has outperformed OpenAI’s o1-mini and Alibaba Group’s QwQ-32B-Preview.
Xiaomi claims that the company’s first major language model was developed using training and reinforcement from a specialized AI task force called Core. Earlier, Chinese media reported that in December last year, the company purchased about 10 thousand GPUs for MiMo training.
Besides, Xiaomi tried to hire Luo Fuli, the developer of another Chinese AI model DeepSeek. However, the girl refused. Meanwhile, Xiaomi’s ambitions include the desire to create a general AI model. Although the company presented MiMo later than DeepSeek and Alibaba, it is convinced that it will have time to catch up with competitors in the race to create AGI.
Meanwhile, OpenAI declare, that their latest AI model, GPT-4.5, released in February, is the first to give the impression of communicating with a thoughtful person.
Major tech companies are investing billions in AI infrastructure. This year, Microsoft, Amazon, Google, and Meta are expected to spend a combined $320 billion in capital expenditures to expand their AI capabilities. For its part, Google has updated the Gemma AI family of «open» models andintroduced Gemma 3— the most powerful AI to date, according to the company, that can be run on a single graphics card.
California-based startup Inception Labs introduced the world’s first large-scale diffusion-based language model developed for commercial use. Diffusion is a technique commonly used by artificial intelligence to generate images and videos. Diffusion works differently — it doesn’t move from left to right, but creates all the text at once. In this case, everything starts with «noise», which is gradually cleaned and a stream of tokens is obtained.
Source: South China Morning Post
Spelling error report
The following text will be sent to our editors: